A White Paper from the Experts in Business-Critical Continuity Energy Logic: Reducing Data Center Energy Consumption by Creating Savings that Cascade Across Systems The model demonstrates Executive Summary that reductions in energy consumption at the IT equipment level have the greatest impact on overall consumption because they cascade across all supporting systems. A number of associations, consultants and vendors have promoted best practices for enhancing data center energy efficiency. These practices cover everything from facility lighting to cooling system design, and have proven useful in helping some companies slow or reverse the trend of rising data center energy consumption. However, most organizations still lack a cohesive, holistic approach for reducing data center energy use. Emerson Network Power analyzed the available energy-saving opportunities and identified the top ten. Each of these ten opportunities were then applied to a 5,000-square-foot data center model based on real-world technologies and operating parameters. Through the model, Emerson Network Power was able to quantify the savings of each action at the system level, as well as identify how energy reduction in some systems affects consumption in supporting systems. The model demonstrates that reductions in energy consumption at the IT equipment level have the greatest impact on overall consumption because they cascade across all supporting systems. This led to the development of Energy Logic, a vendor-neutral roadmap for optimizing data center energy efficiency that starts with the IT equipment and progresses to the support infrastructure. This paper shows how Energy Logic can deliver a 50 percent or greater reduction in data center energy consumption without compromising performance or availability. This approach has the added benefit of removing the three most critical constraints faced by data center managers today: power, cooling and space. In the model, the 10 Energy Logic strategies freed up two-thirds of floor space, one-third of UPS capacity and 40 percent of precision cooling capacity. All of the technologies used in the Energy Logic approach are available today and many can be phased into the data center as part of regular technology upgrades/refreshes, minimizing capital expenditures. The model also identified some gaps in existing technologies that could enable greater energy reductions and help organizations make better decisions regarding the most efficient technologies for a particular data center. 1 Introduction In a report to the U.S. Congress, the Environmental Protection Agency concluded that best practices can reduce data center energy consumption by 50 percent by 2011. The double impact of rising data center energy consumption and rising energy costs has elevated the importance of data center efficiency as a strategy to reduce costs, manage capacity and promote environmental responsibility. Data center energy consumption is being driven by the demand within almost every organization for greater computing capacity The EPA report included a list of Top 10 Energy Saving Best Practices as identified by the Lawrence Berkeley National Lab. Other organizations, including Emerson Network Power, have distributed similar information and there is evidence that some best practices are being adopted. Data center energy consumption has been driven by the demand within almost every organization for greater computing capacity and increased IT centralization. While this was occurring, global electricity prices increased 56 percent between 2002 and 2006. The Spring 2007 DCUG Survey found that 77 percent of respondents already had their data center arranged in a hot-aisle/coldaisle configuration to increase cooling system efficiency, 65 percent use blanking panels to minimize recirculation of hot air and 56 percent have sealed the floor to prevent cooling losses. The financial implications are significant; estimates of annual power costs for U.S. data centers now range as high as $3.3 billion. This trend impacts data center capacity as well. According to the Fall 2007 Survey of the Data Center Users Group (DCUG®), an influential group of data center managers, power limitations were cited as the primary factor limiting growth by 46 percent of respondents, more than any other factor. While progress has been made, an objective, vendor-neutral evaluation of efficiency opportunities across the spectrum of data center systems has been lacking. This has made it difficult for data center managers to prioritize efficiency efforts and tailor best practices to their data center equipment and operating practices. In addition to financial and capacity considerations, reducing data center energy use has become a priority for organizations seeking to reduce their environmental footprint. This paper closes that gap by outlining a holistic approach to energy reduction, based on quantitative analysis, that enables a 50 percent or greater reduction in data center energy consumption. There is general agreement that improvements in data center efficiency are possible. 2 and increased IT centralization. The distinction between percent of total consumption. Supply-side systems include the UPS, power distribution, cooling, lighting and building switchgear, and account for 48 percent of consumption. Data Center Energy Consumption demand and supply power consumption is valuable because reductions in demand-side energy use cascade through the supply side. The first step in prioritizing energy saving opportunities was to gain a solid understanding of data center energy consumption. Emerson Network Power modeled energy consumption for a typical 5,000-squarefoot data center (Figure 1) and analyzed how energy is used within the facility. Energy use was categorized as either “demand side” or “supply side.” Information on data center and infrastructure equipment and operating parameters on which the analysis was based are presented in Appendix A. Note that all data centers are different and the savings potential will vary by facility. However, at minimum, this analysis provides an order-ofmagnitude comparison for data center energy reduction strategies. Demand-side systems are the servers, storage, communications and other IT systems that support the business. Supply-side systems exist to support the demand side. The distinction between demand and supply power consumption is valuable because reductions in demand-side energy use cascade through the supply side. For example, in the 5,000-square-foot data center used to In this analysis, demand-side systems— which include processors, server power supplies, other server components, storage and communication equipment—account for 52 Power and Cooling 48% Computing Equipment 52% Computing Equipment 52% (Demand) Building Switchgear / MV Transformer 3% Lighting 1% Processor 15% Support Systems 48% (Supply) Cooling 38% Server Power Supply 14% UPS 5% Power Draw* Computing 588 kW Lighting 10 kW UPS and distribution losses 72 kW Cooling power draw for computing and UPS losses Other Server 15% PDU 1% Category Building switchgear/MV transformer/other losses Storage 4% TOTAL 429 kW 28 kW 1127 kW Communication Equipment 4% Figure 1. Analysis of a typical 5000-square-foot data center shows that demand-side computing equipment accounts for 52 percent of energy usage and supply-side systems account for 48 percent. * This represents the average power draw (kW). Daily energy consumption (kWh) can be captured by multiplying the power draw by 24. 3 complete. The energy saving measures included in Energy Logic should be considered a guide. Many organizations will already have undertaken some measures at the end of the sequence or will have to deploy some technologies out of sequence to remove existing constraints to growth. analyze energy consumption, a 1 Watt reduction at the server component level (processor, memory, hard disk, etc.) results in an additional 1.84 Watt savings in the power supply, power distribution system, UPS system, cooling system and building entrance switchgear and medium voltage transformer (Figure 2). As these technologies are specified in new equipment, inefficient servers will be phased out and replaced with higherefficiency units, creating a solid foundation for an energy-optimized data center. The Energy Logic Approach Energy Logic takes a sequential approach to reducing energy costs, applying the 10 technologies and best practices that exhibited the most potential in the order in which they have the greatest impact. Power management software has great potential to reduce energy costs and should be considered as part of an energy optimization strategy, particularly for data centers that have large differences between peak and average utilization rates. Other facilities may While the sequence is important, Energy Logic is not intended to be a step-by-step approach in the sense that each step can only be undertaken after the previous one is 1 Watt saved here The Cascade Effect -1.0 W DC-DC Saves an additional .18 W here -1.18 W AC-DC -1.49 W and .31 W here Energy Logic approach is to establish an IT equipment procurement policy that exploits the energy efficiency benefits of low power processors and high-efficiency The first step in the Energy Logic approach is to establish an IT equipment procurement policy that exploits the energy efficiency benefits of low power processors and high-efficiency power supplies. Consequently, every Watt of savings that can be achieved on the processor level creates a total of 2.84 Watts of savings for the facility. Server Component The first step in the Power Distribution and .04 W here -1.53 W UPS and .14 W here 1 Watt saved at the processor saves a total of 2.84 Watts of total consumption -1.67 W Cooling and 1.07 W here -2.74 W Building Switchgear/ Transformer -2.84 W = Cumulative Saving and .07 W here Figure 2. With the Cascade Effect, a 1 Watt savings at the server component level creates a reduction in facility energy consumption of 2.84 Watts. 4 power supplies. Total floor space required for IT equipment was reduced by 65 percent. Reducing Energy Consumption and Eliminating Constraints to Growth choose not to employ power management because of concerns about response times. A significant opportunity exists within the industry to enhance the sophistication of power management to make it an even more powerful tool in managing energy use. Employing the Energy Logic approach to the model data center reduced energy use by 52 percent without compromising performance or availability. The next step involves IT projects that may not be driven by efficiency considerations, but have an impact on energy consumption. They include: • Blade servers • Server virtualization In its unoptimized state, the 5,000-squarefoot data center model used to develop the Energy Logic approach supported a total compute load of 588 kW and total facility load of 1127 kW. Through the optimization strategies presented here, this facility has been transformed to enable the same level of performance using significantly less power and space. Total compute load was reduced to 367 kW, while rack density was increased from 2.8 kW per rack to 6.1 kW per rack. These technologies have emerged as “best practice” approaches to data center management and play a role in optimizing a data center for efficiency, performance and manageability. Once policies and plans have been put in place to optimize IT systems, the focus shifts to supply-side systems. The most effective approaches to infrastructure optimization include: • Cooling best practices • 415V AC power distribution • Variable capacity cooling • Supplemental cooling • Monitoring and optimization This has reduced the number of racks required to support the compute load from 210 to 60 and eliminated power, cooling and space limitations constraining growth (Figure 4). Total energy consumption was reduced to 542 kW and the total floor space required for IT equipment was reduced by 65 percent (Figure 5). Energy Logic is suitable for every type of data center; however, the sequence may be affected by facility type. Facilities operating at high utilization rates throughout a 24-hour day will want to focus initial efforts on sourcing IT equipment with low power processors and high efficiency power supplies. Facilities that experience predictable peaks in activity may achieve the greatest benefit from power management technology. Figure 6 shows how compute load and type of operation influence priorities. Emerson Network Power has quantified the savings that can be achieved through each of these actions individually and as part of the Energy Logic sequence (Figure 3). Note that savings for supply-side systems look smaller when taken as part of Energy Logic because those systems are now supporting a smaller load. 5 Savings Independent of Other Actions Energy Logic Savings with the Cascade Effect Energy Saving Action ROI Savings (kW) Savings (%) Savings (kW) Savings (%) Cumulative Savings (kW) Lower power processors 111 10% 111 10% 111 12 to 18 mo. High-efficiency power supplies 141 12% 124 11% 235 5 to 7 mo. Power management features 125 11% 86 8% 321 Immediate 8 1% 7 1% 328 TCO reduced 38%* Server virtualization 156 14% 86 8% 414 TCO reduced 63%** 415V AC power distribution 34 3% 20 2% 434 2 to 3 mo. Cooling best practices 24 2% 15 1% 449 4 to 6 mo. Variable capacity cooling: variable speed fan drives 79 7% 49 4% 498 4 to 10 mo. Supplemental cooling 200 18% 72 6% 570 10 to 12 mo. Monitoring and optimization: Cooling units work as a team 25 2% 15 1% 585 3 to 6 mo. Blade servers * Source for blade impact on TCO: IDC ** Source for virtualization impact on TCO: VMware Figure 3. Using the model of a 5,000-square-foot data center consuming 1127 kW of power, the actions included in the Energy Logic approach work together to produce a 585 kW reduction in energy use. Constraint Unoptimized Optimized Capacity Freed Up 4988 1768 3220 (65%) 2 * 750 2 * 500 2 * 250 (33%) Cooling plant capacity (tons) 350 200 150 (43%) Building entrance switchgear and genset (kW) 1169 620 549 (47%) Data center space (sq ft) UPS capacity (kVA) Figure 4. Energy Logic removes constraints to growth in addition to reducing energy consumption. 6 P2 P1 P2 P1 D1 R R R R R R R R R R D2 D1 R R R R R R R R D2 D1 R R R R HOT AISLE R R R CW CRAC R R D1 R R R R R R R R R R R R D2 D1 R R R R R R R R D2 D1 R R R R CW CRAC R R D1 R R R R R R R R R R R R D2 D1 R R R R R R R R D2 D1 R R R R CW CRAC R R R R R R D2 D1 R R R R R R R R R R D2 D1 R R R R R R R R R R R R R R R R R R R R D2 D1 R R R R R R R R R R R R D2 R R R R R R D2 R R R D2 R R R D2 R R R D2 R R R D2 R R R D2 D1 R R R R R R R D2 D1 R R R R R R R R R R D2 HOT AISLE R R R Battery Room UPS Room COLD AISLE R R R D2 D1 R R R R R R R D2 D1 R R R R R R R HOT AISLE R Control Center COLD AISLE COLD AISLE HOT AISLE R HOT AISLE HOT AISLE COLD AISLE D1 R R COLD AISLE HOT AISLE R R HOT AISLE COLD AISLE D1 R R COLD AISLE HOT AISLE R R HOT AISLE COLD AISLE D1 R R R R R COLD AISLE R R D2 D1 R R R HOT AISLE R R R R HOT AISLE CW CRAC R CW CRAC R COLD AISLE CW CRAC D1 R COLD AISLE CW CRAC CW CRAC COLD AISLE P2 CW CRAC P1 Before Energy Logic P1 P2 P1 P2 R R R D2 D1 R D2 D1 XDV R R R R COLD AISLE D2 D1 XDV R R R R R D2 D2 D2 HOT AISLE XDV R XDV HOT AISLE D1 R R Control Center COLD AISLE XDV R XDV R XDV R XDV CW CRAC D1 R R D2 CW CRAC R R XDV R R XDV R R XDV R XDV D1 R R HOT AISLE COLD AISLE Space Savings: 65% R CW CRAC R D2 D1 R XDV R XDV R XDV R HOT AISLE XDV R XDV R XDV R XDV D1 R COLD AISLE XDV CW CRAC COLD AISLE Battery Room COLD AISLE CW CRAC UPS Room After Energy Logic P1 = Stage one distribution, side A D1 = Stage two distribution, side A P2 = Stage one distribution, side B D2 = Stage two distribution, side B CW CRAC = Chilled water computer room air conditioner R = Rack XDV = Supplemental cooling module Figure 5. The top diagram shows the unoptimized data center layout. The lower diagram shows the data center after the Energy Logic actions were applied. Space required to support data center equipment was reduced from 5,000 square feet to 1,768 square feet (65 percent). 7 1. Lowest power processor 2. High-efficiency power supplies 3. Blade servers 4. Power management Daytime operations 24-hour operations 1. Virtualization 2. Lowest power processor 3. High-efficiency power supply 4. Power management 5. Consider mainCooling best practices frame architecture Variable capacity cooling Supplemental cooling 1. Virtualization 415V AC distribution Monitoring and optimization 2. Power management 1. Power management 2. Low power processor 3. High-efficiency power supplies 4. Blade servers 3. Low power processor 4. High-efficiency power supplies Compute Intensive I/O Intensive Figure 6 . The Energy Logic approach can be tailored to the compute load and type of operation. The Ten Energy Logic Actions lower power processors deliver the same performance as higher power models (Figure 8). 1. Processor Efficiency In the absence of a true standard measure of processor efficiency comparable to the U.S. Department of Transportation fuel efficiency standard for automobiles, Thermal Design Power (TDP) serves as a proxy for server power consumption. In the 5,000-square-foot data center modeled for this paper, low power processors create a 10 percent reduction in overall data center power consumption. 2. Power Supplies As with processors, many of the server power supplies in use today are operating at efficiencies below what is currently available. The U.S. EPA estimated the average efficiency of installed server power supplies at 72 percent in 2005. In the model, we assume the unoptimized data center uses power supplies that average 79 percent across a mix of servers that range from four-years old to new. The typical TDP of processors in use today is between 80 and 103 Watts (91 W average). For a price premium, processor manufacturers provide lower voltage versions of their processors that consumes on average 30 Watts less than standard processors (Figure 7). Independent research studies show these Sockets Speed (GHz) Standard Low power Saving 1 1.8-2.6 103 W 65 W 38 W 2 1.8-2.6 95 W 68 W 27 W 2 1.8-2.6 80 W 50 W 30 W AMD Intel Figure 7. Intel and AMD offer a variety of low power processors that deliver average savings between 27 W and 38 W. 8 AS3AP System Performance Results Transaction/Second (Higher is Better) Best-in-class power supplies are available today that deliver efficiency of 90 percent. Use of these power supplies reduces power draw within the data center by 124 kW or 11 percent of the 1127 kW total. 30000 25000 As with other data center systems, server power supply efficiency varies depending on load. Some power supplies perform better at partial loads than others and this is particularly important in dual-corded devices where power supply utilization can average less than 30 percent. Figure 9 shows power supply efficiencies at different loads for two power supply models. At 20 percent load, model A has an efficiency of approximately 88 percent while model B has an efficiency closer to 82 percent. Transactions/Second 20000 15000 10000 5000 0 Load 1 Load 2 Opteron 2218 HE 2.6 GHz Load 3 Woodcrest LV S148 2.3 GHz Load 4 Opteron 2218 2.6 GHz Load 5 Woodcrest 5140 2.3 GHz Source Anandtech Figure 8. Performance results for standard- and low-power processor systems using the American National Standard Institute AS3AP benchmark. Figure 9 also highlights another opportunity to increase efficiency: sizing power supplies closer to actual load. Notice that the maximum configuration is about 80 percent of “Typical” PSU Load in Redundant Config. “Typical” Config 100% CPU Util. “Max” Configuration Nameplate Rating 90% Efficiency 85% 80% Model A Model B 75% 70% 0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 50% 55% 60% 65% 70% 75% 80% 85% 90% 95% 100% Percent of Full Load Figure 9. Power supply efficiency can vary significantly depending on load and power supplies are often sized for a load that exceeds the maximum server configuration. 9 While the move to blade are disabled because of concerns regarding response time; however, this decision may need to be reevaluated in light of the significant savings this technology can enable. the nameplate rating and the typical configuration is 67 percent of the nameplate rating. Server manufacturers should allow purchasers to choose power supplies sized for a typical or maximum configuration. servers is typically not driven by energy considerations, blade servers can play a role in reduc- In the model, we assume that the idle power draw is 80 percent of the peak power draw without power management, and reduces to 45 percent of peak power draw as power management is enabled. With this scenario, power management can save an additional 86 kW or eight percent of the unoptimized data center load. 3. Power management software Data centers are sized for peak conditions that may rarely exist. In a typical business data center, daily demand progressively increases from about 5 a.m. to 11 a.m. and then begins to drop again at 5 p.m. (Figure 10). Server power consumption remains relatively high as server load decreases (Figure 11). In idle mode, most servers consume between 70 and 85 percent of full operational power. Consequently, a facility operating at just 20 percent capacity may use 80 percent of the energy as the same facility operating at 100 percent capacity. 4. Blade Servers Many organizations have implemented blade servers to meet processing requirements and improve server management. While the move to blade servers is typically not driven by energy considerations, blade servers can play a role in energy consumption. Server processors have power management features built-in that can reduce power when the processor is idle. Too often these features Blade servers consume about 10 percent less power than equivalent rack mount servers because multiple servers share common Servers ing energy consumption. AC Power Input Versus Percent CPU Time 45 150 40 120 Dell Power Edge 2400 (Web/SQL Server) 125 35 Watts 100 30 100 80 75 60 50 40 25 20 0 0 20 15 10 % CPU Time 25 Watts Percent Utilization Proc Time % 5 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Hours Figure 10. Daily utilization for a typical business data center. Figure 11. Low processor activity does not translate into low power consumption. 10 Implementing virtualization provides an data center airflow (Figure 12). Many organizations, including Emerson Network Power, offer CFD imaging as part of data center assessment services focused on improving cooling efficiency. power supplies, cooling fans and other components. incremental eight percent reduction in total data center power draw for the 5,000-squarefoot facility. In the model, we see a one percent reduction in total energy consumption when 20 percent of rack-based servers are replaced with blade servers. More importantly, blades facilitate the move to a high-density data center architecture, which can significantly reduce energy consumption. Additionally, temperatures in the cold aisle may be able to be raised if current temperatures are below 68° F. Chilled water temperatures can often be raised from 45° F to 50° F. 5. Server Virtualization As server technologies are optimized, virtualization is increasingly being deployed to increase server utilization and reduce the number of servers required. In the model, cooling system efficiency is improved five percent simply by implementing best practices. This reduces overall facility energy costs by one percent with virtually no investment in new technology. In our model, we assume that 25 percent of servers are virtualized with eight non-virtualized physical servers being replaced by one virtualized physical server. We also assume that the applications being virtualized were residing in single-processor and two-processor servers and the virtualized applications are hosted on servers with at least two processors. 7. 415V AC Power Distribution The critical power system represents another opportunity to reduce energy consumption; however, even more than other systems, care must be taken to ensure reductions in energy consumption are not achieved at the cost of reduced equipment availability. Most data centers use a type of UPS called a double-conversion system. These systems convert incoming power to DC and then back to AC within the UPS. This enables the UPS to generate a clean, consistent waveform for IT equipment and effectively isolates IT equipment from the power source. Implementing virtualization provides an incremental eight percent reduction in total data center power draw for the 5,000-squarefoot facility. 6. Cooling Best Practices Most data centers have implemented some best practices, such as the hot-aisle/cold-aisle rack arrangement. Potential exists in sealing gaps in floors, using blanking panels in open spaces in racks, and avoiding mixing of hot and cold air. ASHRAE has published several excellent papers on these best practices. UPS systems that don’t convert the incoming power—line interactive or passive standby systems—can operate at higher efficiencies because of the losses associated with the conversion process. These systems may compromise equipment protection because they do not fully condition incoming power. Computational fluid dynamics (CFD) can be used to identify inefficiencies and optimize A bigger opportunity exists downstream from the UPS. In most data centers, the UPS 11 8. Variable Capacity Cooling Data center systems are sized to handle peak loads, which rarely exist. Consequently, operating efficiency at full load is often not a good indication of actual operating efficiency. Newer technologies, such as Digital Scroll compressors and variable frequency drives in computer room air conditioners (CRACs), allow high efficiencies to be maintained at partial loads. Figure 12. CFD imaging can be used to evaluate cooling efficiency and optimize airflow. This image shows hot air being recirculated as it is pulled back toward the CRAC, which is poorly positioned. Digital scroll compressors allow the capacity of room air conditioners to be matched exactly to room conditions without turning compressors on and off. provides power at 480V, which is then stepped down via a transformer, with accompanying losses, to 208V in the power distribution system. These stepdown losses can be eliminated by converting UPS output power to 415V. Typically, CRAC fans run at a constant speed and deliver a constant volume of air flow. Converting these fans to variable frequency drive fans allows fan speed and power draw to be reduced as load decreases. Fan power is directly proportional to the cube of fan rpm and a 20 percent reduction in fan speed provides almost 50 percent savings in fan power consumption. These drives are available in retrofit kits that make it easy to upgrade existing CRACs with a payback of less than one year. The 415V three-phase input provides 240V single-phase, line-to-neutral input directly to the server (Figure 13). This higher voltage not only eliminates stepdown losses but also enables an increase in server power supply efficiency. Servers and other IT equipment can handle 240V AC input without any issues. In the model, an incremental two percent reduction in facility energy use is achieved by using 415V AC power distribution. Traditional 208V Distribution 480V UPS 480V Servers Storage Comm Equipment 12V PDU 208V transformer PDU panelboard 208V PSU 415V 3-phase 240V phase to neutral PDU panelboard 240V PSU 415V Distribution 480V UPS Servers Storage Comm Equipment 12V Figure 13. 415V power distribution provides a more efficient alternative to using 208V power. 12 Newer technologies such as Digital Scroll compressors and variable frequency drives in computer room air conditioners (CRACs) allow high efficiencies to be maintained at partial loads. Supplemental cooling units are mounted above or alongside equipment racks and pull hot air directly from In the chilled water-based air conditioning system used in this analysis, the use of variable frequency drives provides an incremental saving of four percent in data center power consumption. the hot aisle and deliver cold air to the cold aisle. 9. High Density Supplemental Cooling Traditional room-cooling systems have proven very effective at maintaining a safe, controlled environment for IT equipment. However, optimizing data center energy efficiency requires moving from traditional data center densities (2 to 3 kW per rack) to an environment that can support much higher densities (in excess of 30 kW). This requires implementing an approach to cooling that shifts some of the cooling load from traditional CRAC units to supplemental cooling units. Supplemental cooling units are mounted above or alongside equipment racks (Figure14), and pull hot air directly from the hot aisle and deliver cold air to the cold aisle. Supplemental cooling units can reduce cooling costs by 30 percent compared to traditional approaches to cooling. These savings are achieved because supplemental cooling brings cooling closer to the source of heat, reducing the fan power required to move air. They also use more efficient heat exchangers and deliver only sensible cooling, which is ideal for the dry heat generated by electronic equipment. Refrigerant is delivered to the supplemental cooling modules through an overhead piping system, which, once installed, allows cooling modules to be easily added or relocated as the environment changes. Figure 14. Supplemental cooling enables higher rack densities and improved cooling efficiency. In the model, 20 racks at 12 kW density per rack use high density supplemental cooling while the remaining 40 racks (at 3.2 kW density) are supported by the traditional room cooling system. This creates an incremental six percent reduction in overall data center energy costs. As the facility evolves and more racks move to high density, the savings will increase. 10. Monitoring and Optimization One of the consequences of rising equipment densities has been increased diversity within the data center. Rack densities are rarely uniform across a facility and this can create cooling inefficiencies if monitoring and optimization is not implemented. Room cooling units on one side of a facility may be humidifying the environment based on local conditions while units on the opposite side of the facility are dehumidifying. Cooling control systems can monitor conditions across the data center and coordinate the activities of multiple units to prevent conflicts and increase teamwork (Figure 15). 13 TEAMWORK MODE Humidifier ON Unit 1 Humidifier ON Unit 2 UNBALANCED LOAD Control Disables heaters in “Cold Zone” Compressors ON Dehumidification ON Unit 3 Unit 1 SWITCH OR HUB NO LOAD Unit 2 Unit 3 UNBALANCED LOAD NO LOAD Unit 5 Unit 4 POWER Unit 6 Unit 5 Humidifier ON Unit 4 Unit 6 Compressors ON Dehumidification ON Figure 15. System-level control reduces conflict between room air conditioning units operating in different zones in the data center. In the model, an incremental saving of one percent is achieved as a result of system-level monitoring and control. • Monitor and reduce parasitic losses from generators, exterior lighting and perimeter access control. For a 1 MW load, generator losses of 20 kW to 50 kW have been measured. Other Opportunities for Savings The Energy Logic model also provides a roadmap for the industry to deliver the information and technologies that data center managers can use Energy Logic prioritizes the most effective energy reduction strategies, but it is not intended to be a comprehensive list of energy reducing measures. In addition to the actions in the Energy Logic strategy, data center managers should consider the feasibility of the following: What the Industry Can Learn from the Energy Logic Model The Energy Logic model not only prioritizes actions for data center managers. It also provides a roadmap for the industry to deliver the information and technologies that data center managers can use to optimize their facilities. Here are specific actions the industry can take to support increased data center efficiency: • Consolidate data storage from direct attached storage to network attached storage. Also, faster disks consume more power so consider reorganizing data so that less frequently used data is on slower archival drives. 1. Define universally accepted metrics for processor, server and data center efficiency There have been tremendous technology advances in server processors in the last decade. Until 2005, higher processor performance was linked with higher clock speeds and hotter chips consuming more power. Recent advances in multi-core technology have driven performance increases by using more computing cores operating at relatively lesser clock speeds, which reduces power consumption. • Use economizers where appropriate to allow outside air to be used to support data center cooling during colder months, creating opportunities for energy-free cooling. With today’s high-density computing environment, economizers can be cost effective in many more locations than might be expected. 14 to optimize their facilities. An industry standard of data center efficiency that measures performance per Watt of energy used would be extremely beneficial in measuring the progress of data center optimization efforts. Today processor manufacturers offer a range of server processors from which a customer needs to select the right processor for the given application. What is lacking is an easyto-understand and easy-to-use measure such as the miles-per-gallon automotive fuel efficiency ratings developed by the U.S. Department of Transportation, that can help buyers select the ideal processor for a given load. The performance per Watt metric is evolving gradually with SPEC score being used as the server performance measure, but more work is needed. eliminate much of the resistance to using power management. 3. Matching power supply capacity to server configuration Server manufacturers tend to oversize power supplies to accommodate the maximum configuration of a particular server. Some users may be willing to pay an efficiency penalty for the flexibility to more easily upgrade, but many would prefer a choice between a power supply sized for a standard configuration and one sized for maximum configuration. Server manufacturers should consider making these options available and users need to be educated about the impact power supply size has on energy consumption. This same philosophy could be applied to the facility level. An industry standard of data center efficiency that measures performance per Watt of energy used would be extremely beneficial in measuring the progress of data center optimization efforts. The PUE ratio developed by the Green Grid provides a measure of infrastructure efficiency, but not total facility efficiency. IT management needs to work with IT equipment and infrastructure manufacturers to develop the miles-per-gallon equivalent for both systems and facilities. 4. Designing for high density A perception persists that high-density environments are more expensive than simply spreading the load over a larger space. Highdensity environments employing blade and virtualized servers are actually economical as they drive down energy costs and remove constraints to growth, often delaying or eliminating the need to build new facilities. 2. More sophisticated power management While enabling power management features provides tremendous savings, IT management often prefers to stay away from this technology as the impact on availability is not clearly established. As more tools become available to manage power management features, and data is available to ensure that availability is not impacted, we should see this technology gain market acceptance. More sophisticated controls that would allow these features to be enabled only during periods of low utilization, or turned off when critical applications are being processed, would 5. High-voltage distribution 415V power distribution is used commonly in Europe, but UPS systems that easily support this architecture are not readily available in the United States. Manufacturers of critical power equipment should provide the 415V output as an option on UPS systems and can do more to educate their customers regarding high-voltage power distribution. 6. Integrated measurement and control Data that can be easily collected from IT systems and the racks that support them has 15 yet to be effectively integrated with support systems controls. This level of integration would allow IT systems, applications and support systems to be more effectively managed based on actual conditions at the IT equipment level. IT consolidation projects also play an important role in data center optimization. Both blade servers and virtualization contribute to energy savings and support a high-density environment that facilitates true optimization. Conclusion The final steps in the Energy Logic optimization strategy is to focus on infrastructure systems, employing a combination of best practices and efficient technologies to increase the efficiency of power and cooling systems. Together these strategies created a 52 percent reduction in energy use in the 5,000-squarefoot data center model developed by Emerson Network Power while Data center managers and designers, IT equipment manufacturers and infrastructure providers must all collaborate to truly optimize data center efficiency. Together these strategies created a 52 percent reduction in energy use in the 5,000square-foot data center model developed by Emerson Network Power while removing constraints to growth. For data center managers, there are a number of actions that can be taken today that can significantly drive down energy consumption while freeing physical space and power and cooling capacity to support growth. Appendix B shows exactly how these savings are achieved over time as legacy technologies are phased out and savings cascade across systems. Energy reduction initiatives should begin with policies that encourage the use of efficient IT technologies, specifically low power processors and high-efficiency power supplies. This will allow more efficient technologies to be introduced into the data center as part of the normal equipment replacement cycle. Power management software should also be considered in applications where it is appropriate as it may provide greater savings than any other single technology, depending on data center utilization. 16 removing constraints to growth. Appendix A: Data Center Assumptions Used to Model Energy Use To quantify the results of the efficiency improvement options presented in this paper, a hypothetical data center that has not been optimized for energy efficiency was created. The ten efficiency actions presented in this paper were then applied to this facility sequentially to quantify results. Servers • Age is based on average server replacement cycle of 4-5 years. • Processor Thermal Design Power averages 91W/processor. • All servers have dual redundant power supplies. The average DC-DC conversion efficiency is assumed at 85% and average AC-DC conversion efficiency is assumed at 79 percent for the mix of servers from four-years old to new. This 5000-square-foot hypothetical data center has 210 racks with average heat density of 2.8kW/rack. The racks are arranged in a hot-aisle/cold-aisle configuration. Cold aisles are four feet wide, and hot aisles are three feet wide. Based on this configuration and operating parameters, average facility power draw was calculated to be 1127 kW. Following are additional details used in the analysis. • Daytime power draw is assumed to exist for 14 hours on weekdays and 4 hours on weekends. Night time power draw is 80 percent of daytime power draw. • See Figure 16 for more details on server configuration and operating parameters. Single Socket Two sockets Four Sockets More than four Total Number of servers 157 812 84 11 1064 Daytime power draw (Watts/server) 277 446 893 4387 – Nighttime power draw (Watts/server) 247 388 775 3605 – Total daytime power draw (kW) 44 362 75 47 528 Total nighttime power draw (kW) 39 315 65 38 457 Average server power draw (kW) 41 337 70 42 490 Figure 16. Server operating parameters used in the Energy Logic model. 17 Storage Cooling system • Storage Type: Network attached storage. • Cooling System is chilled water based. • Capacity is 120 Terabytes. • Total sensible heat load on the precision cooling system includes heat generated by the IT equipment, UPS and PDUs, building egress and human load. • Average Power Draw is 49 kW. Communication Equipment • Routers, switches and hubs required to interconnect the servers, storage and access points through Local Area Network and provide secure access to public networks. • Cooling System Components: - Eight 128.5 kW chilled water based precision cooling system placed at the end of each hot aisle. Includes one redundant unit. - The chilled water source is a chiller plant consisting of three 200 ton chillers (n+1) with matching condensers for heat rejection and four chilled water pumps (n+2). • Average Power Draw is 49 kW. Power Distribution Units (PDU): • Provides output of 208V, 3 Phase through whips and rack power strips to power servers, storage, communication equipment and lighting. (Average load is 539kW). • Input from UPS is 480V 3-phase. - The chiller, pumps and air conditioners are powered from the building distribution board (480V 3 phase). • Efficiency of power distribution is 97.5 percent. - Total cooling system power draw is 429 kW. UPS System Building substation: • Two double conversion 750 kVA UPS modules arranged in dual redundant (1 +1) configuration with input filters for power factor correction (power factor = 91 percent). • The building substation provides 480V 3-phase power to UPS’s and cooling system. • Average load on building substation is 1,099 kW. • Utility input is 13.5 kVA, 3-phase connection. • System consists of transformer with isolation switchgear on the incoming line, switchgear, circuit breakers and distribution panel on the low voltage line. • Substation, transformer and building entrance switchgear composite efficiency is 97.5 percent. • The UPS receives 480V input power for the distribution board and provides a 480V, 3 Phase power to the power distribution units on the data center floor. • UPS efficiency at part load: 91.5 percent. 18 Appendix B: Timing Benefits from Efficiency Improvement Actions Infrastructure Projects Best Practices IT Projects IT Policies Efficiency Improvement Area Estimated Cumulative Yearly Savings Savings (kW) Year 1 Year 2 Year 3 Year 4 Year 5 Low power processors 111 6 22 45 78 111 Higher efficiency power supplies 124 12 43 68 99 124 Server power management 86 9 26 43 65 86 Blade servers 7 1 7 7 7 7 Virtualization 86 9 65 69 86 86 415V AC power distribution 20 0 0 20 20 20 Cooling best practices 15 15 15 15 15 15 Variable capacity cooling 49 49 49 49 49 49 High density supplemental cooling 72 0 57 72 72 72 Monitoring and optimization 15 0 15 15 15 15 585 100 299 402 505 585 19 Note: Energy Logic is an approach developed by Emerson Network Power to provide organizations with the information they need to more effectively reduce data center energy consumption. It is not directly tied to a particular Emerson Network Power product or service. We encourage use of the Energy Logic approach in industry discussions on energy efficiency and will permit use of the Energy Logic graphics with the following attribution: Emerson Network Power 1050 Dearborn Drive P.O. Box 29186 Columbus, Ohio 43229 800.877.9222 (U.S. & Canada Only) 614.888.0246 (Outside U.S.) Fax: 614.841.6022 EmersonNetworkPower.com Liebert.com The Energy Logic approach developed by Emerson Network Power. While every precaution has been taken to ensure accuracy and completeness in this literature, Liebert Corporation assumes no responsibility, and disclaims all liability for damages resulting from use of this information or for any errors or omissions. Specifications subject to change without notice. © 2007 Liebert Corporation. All rights reserved throughout the world. Trademarks or registered trademarks are property of their respective owners. ® Liebert and the Liebert logo are registered trademarks of the Liebert Corporation Business-Critical Continuity, Emerson Network Power and the Emerson Network Power logo are trademarks and service mark of the Emerson Electric Co. To request a graphic send an email to energylogic@emerson.com. WP154-158-117 SL-24621 Emerson Network Power. The global leader in enabling Business-Critical Continuity™. EmersonNetworkPower. com AC Power Embedded Computing Outside Plant Racks & Integrated Cabinets Connectivity Embedded Power Power Switching & Controls Services DC Power Monitoring Precision Cooling Precision Cooling